Presenter-Centric Image Collection and Annotation: Enhancing Accessibility for the Visually Impaired

With the recent COVID-19 pandemic, our daily lives and workplace became more dependent on technologies, e.g., the intensive use of video conferences. While a highly connected world enables remote working, it also raises new barriers for an already excluded community, the visually impaired people. Low or no eyesight prevents those people from capturing visual infor-mation, which makes it difficult to understand the overall context of a remote presentation. Most Artificial Intelligence methods are specific to sighted-people data domains due to the scarcity of datasets for the visually impaired. In this paper, we propose an approach to collect data automatically and a protocol to annotate this data specifically for this audience, aiming to support the development of Assistive Technology systems. We demonstrate the viability of the proposed methods by creating a dataset and evaluating its quality, diversity, and representativeness through analytical methods and machine learning models.